A two-layer area-level edge detector is proposed in this paper. The first is a detection layer, and the second is a location layer. The proposed approach takes an n×n area as a processing unit. In the detection layer, based on the property of an entropy value, we judge if the square area is a homogeneous region. If the answer is no, then we enter the location layer to locate the edge points if they exist. In the real world, the gray value of a step edge point always lies between those of its two neighboring regions, each with nearly constant gray values. Based on this fact and the moment-preserving principle, a new approach is provided in the layer to locate all of the edge points deterministically. The proposed method releases the assumption, which is usually taken by general area-level detectors, that all of the edge points in a detection area must lie on a line or on a given type of curve. The detected results are noiseless and thin. Two applications of the new edge detector are provided and are compared with the Sobel and the Tabatabai-Mitchell detectors to show their effectiveness.